Hmmm, we can already make self-replicating programs that are incredibly more efficient.

So the shock value of the title depends on convincing readers that LLMs are the stronger definition of "AI", instead of something more like an "iteratively try Stackoverflow answers" system.